Self Improving Agent
PassAudited by ClawScan on May 10, 2026.
Overview
The skill is transparent and purpose-aligned, but it intentionally creates persistent learning/memory context and optional hooks that can influence future agent sessions.
This appears safe to install if you want a persistent learning log for your agent. Before enabling hooks or promoting learnings, review the stored entries for secrets, private details, and incorrect instructions, because these files may influence future agent behavior.
Findings (3)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
A mistaken or sensitive learning could be reused in future sessions and change how the agent behaves.
The skill intentionally promotes learnings into files that future agents may treat as context or instructions. This is purpose-aligned, but persistent memory can preserve incorrect or sensitive content if not reviewed.
Broadly applicable learning | Promote to `CLAUDE.md`, `AGENTS.md`, and/or `.github/copilot-instructions.md`
Review `.learnings/` and promoted files before relying on them; avoid storing secrets, private user data, or unverified instructions.
Information from one session could be read or shared with another session if these tools are used.
The skill documents cross-session transcript reading, messaging, and sub-agent spawning as ways to share learnings. The artifacts do not show automatic use, but users should understand the data boundary.
sessions_history — Read another session's transcript ... sessions_send — Send a learning to another session ... sessions_spawn — Spawn a sub-agent for background work
Use cross-session tools only when needed, avoid sharing sensitive transcript content, and bound any spawned sub-agent task clearly.
If enabled, hook scripts will run automatically during agent sessions.
Optional hook setup runs local shell scripts on prompt submission or Bash tool completion. The provided scripts only emit reminders or scan tool output for error strings, but hooks execute with the agent user's local permissions.
"command": "./skills/self-improvement/scripts/activator.sh" ... "command": "./skills/self-improvement/scripts/error-detector.sh"
Enable hooks only if you want automatic reminders, keep the scripts reviewed, and disable the hook configuration if it becomes intrusive.